Specific emitter identification (SEI) authenticates wireless devices at the physical layer by exploiting device-specific radio frequency fingerprints, offering a lightweight complement to cryptographic identity for Internet-of-Things (IoT) security. In real IoT deployments, new devices join the network over time, while only a few labeled samples are available to register each one, and previously enrolled devices must still be recognized reliably. We study SEI under this few-shot class-incremental setting and propose PrototypeStabilized Adaptive Calibration (PSAC). PSAC constructs stabilized novelclass prototypes through multi-view estimation and neighbor-guided calibration, adapts the decision layer using pseudo old-class replay, lassifier-head refinement, and a support-feature memory, and applies replay-guided consolidation with session-wise score debiasing to limit accumulated drift. Experiments on a public LoRa RF-fingerprint dataset and a self-collected AIS dataset show consistent gains. On the 25-class LoRa protocol, PSAC achieves the best average accuracy of 82.16% and the best final-session accuracy of 70.96%, clearly improving over representative baselines. On the 100-class AIS protocol, PSAC attains an average accuracy of 82.28% with a performance drop of 9.58%, remaining competitive in a larger maritime deviceidentification scenario. Ablation, shot-setting, and hyperparameter analyses further confirm the stability of PSAC under few-shot class-incremental SEI.
RF signal classification provides modulation information that can support wireless spectrum awareness after signal acquisition and segmentation. This paper presents a self-weighted convolutional neural network for classifying communication-signal modulation formats from in-phase/quadrature (IQ) sequences. In the proposed multi-branch block, the normalized output channels of one convolutional branch are multiplied by learnable scaling coefficients before aggregation. Stacked blocks are followed by squeeze-and-excitation recalibration and a fully connected classification head. The network is evaluated on a MATLAB-generated 12-class IQ dataset under variations in network depth, symbol rate, carrier-frequency offset, and transfer settings, and is compared with Inception, ResNet50, CaiT, and ViT for Small-Size Datasets. An institute-provided IQ dataset with the same modulation categories provides an additional within-dataset evaluation. On the synthetic dataset, the proposed network achieves an average test accuracy of 91.91%, 0.52 percentage points above Inception. Fine-tuning on a small target-domain subset achieves 91.81% average accuracy while reducing the observed convergence requirement from 21 to 5 epochs compared with training from scratch on the target dataset. On the institute-provided dataset, the proposed, transfer, and Inception networks achieve 93.61%, 92.75%, and 92.32%, respectively. These results support the proposed method as a modulation-aware RF signal-classification component for wireless spectrum-awareness workflows under the evaluated synthetic and institute-provided data settings.